Estimating the building based energy consumption as an anthropogenic contribution to urban heat islands

Sustainable Cities and Society 19 (2015) 373–384

Contents lists available at ScienceDirect

Sustainable Cities and Society journal homepage: www.elsevier.com/locate/scs

Estimating the building based energy consumption as an anthropogenic contribution to urban heat islands Peter Boehme a , Matthias Berger b,c,∗ , Tobias Massier a a b c

TUM CREATE, 1 CREATE Way, #10-02 CREATE Tower, 138602 Singapore, Singapore Department of Architecture, ETH Zurich, Building HIT, Wolfgang-Pauli-Str. 27, 8093 Zurich, Switzerland Future Cities Laboratories, Singapore-ETH Centre, 1 CREATE Way, #06-01 CREATE Tower, 138602 Singapore, Singapore

a r t i c l e

i n f o

Article history: Available online 27 May 2015 Keywords: Urban heat island Anthropogenic heat Building electricity consumption

a b s t r a c t Today the implication of buildings’ electricity demand on the outdoor climate around buildings is not fully understood. For tropical cities like Singapore, where air-conditioning is required throughout the year and high rise buildings irregularly alternate with lower buildings, the distribution of this so called anthropogenic heat emissions in time and space is determining the local and overall contribution to the urban heat island (UHI). In the absence of detailed measurements calculating the local consumption can be challenging. We present a methodology which combines a top-down disaggregation of sectoral electricity consumption with a bottom-up geographic information system (GIS) derived building database for obtaining anthropogenic heat emission maps with high spatial resolution. The database has been validated through control samples. Using the example of Singapore we can show that heat emissions are more inhomogeneous and higher in magnitude than previously estimated. Our method can now be employed to generate better UHI models by identifying areas with significant anthropogenic heat emissions. © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction Rejected or waste heat from electricity consumption in the residential and commercial building sector is an anthropogenic contribution to the urban heat island (UHI) phenomenon. The heterogeneous city-wide spatial distribution of these heat emissions leads to locally different magnitudes (in W/m2 ). In contrast, the natural insolation is homogeneous, on an urban scale and averaged in time, hence not location dependent. So far, UHI studies have understood anthropogenic heat as a significant but small contribution to the UHI, and distribution patterns of anthropogenic heat have been seen as homogenous with respect to the modelling scale (down to 1 km grid cell size) (Roth, 2007; Quah & Roth, 2012; Oke, 1973, 1982). In our case study of Singapore and for tropical, high-dense, and vertical cities in general we challenge the existing UHI models due to the fact that they have been developed based on lower density and shallow cities (Kolokotroni, Davies, Croxford, Bhuiyan, & Mavrogianni, 2010; Kolokotroni, Zhang, &

∗ Corresponding author at: Singapore-ETH Centre, Future Cities Laboratories, 1 CREATE Way, #06-01 CREATE Tower, Singapore 138602, Singapore. Tel.: +65 9384 9813. E-mail address: [email protected] (M. Berger).

Watkins, 2007; Mavrogianni et al., 2011). Two missing gaps in our understanding of anthropogenic heat contributing to UHI have to be closed: How does the current microclimate modeling approach depend on the morphologies of their cities of origin? Assuming that anthropogenic heat and heterogeneous morphology have a significant role, how can we identify heat sources when limited building-based information is available? In the subsequent analysis we are limiting the scope to buildings and QB from the total anthropogenic heat given by QF = QV + QB + QM , not studying metabolic heat QM and the effect of vehicles QV . 1.1. Urban heat island models Following Kolokotroni et al. (2010), Kolokotroni et al. (2007) urban air temperatures and consequently UHI can be modeled and predicted in four different ways: by means of climatology models, empirical models, computational fluid dynamics models (CFD), and statistical regression methods. Climatology models, as they are used in the case of London within the LUCID project, usually integrate anthropogenic heat emission into the energy balance in a grid of 1 km resolution, where the sub-grid scale is partitioned into land use classes (Mavrogianni et al., 2011). The same has been done under different climatic boundary conditions for Singapore (Li et al., 2013).

http://dx.doi.org/10.1016/j.scs.2015.05.006 2210-6707/© 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4. 0/).

374

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

Fig. 1. “Rooftops of Athens” via Flickr by stanjourdan, April 11, 2013, Creative Commons Attribution.

On a smaller scale, urban canyons are often evaluated by empirical methods taking building and canyon geometry into account (Erell & Williamson, 2006; Hu, Yu, Chen, Li, & Liu, 2012). All these models assume certain homogeneity inside the smallest scale, in terms of geometry of building blocks, spatial distribution of anthropogenic heat emissions and temporal dependencies. Anthropogenic heat is most of the time considered inside the models, yet there is disagreement among experts about the magnitude of heat contributing to the UHI (Quah & Roth, 2012). Important to mention here is the spatiotemporal pattern of heat emission, which is averaged as mean annual values of heat flux densities in W/m2 /a (Zhou, Weng, Gurney, Shuai, & Hu, 2012), or alternatively as annual energy densities in kW h/m2 /a (Rode, Keim, Robazza, Viejo, & Schofield, 2014). It arises naturally that the smaller the spatial grid size is, the higher the peak densities are. Values are ranging for instance in London from 9 W/m2 for the entire city to 150 W/m2 peak for a 1 km grid to 210 W/m2 for a 200 m grid (Hamilton et al., 2009; Kotthaus & Grimmond, 2012). Hamilton et al. investigated the relation of individual buildings’ heat emissions to the insolation. Based on a daily total, the insolation exceeded the anthropogenic heat by factor 23 on a warm and sunny summer day. However, a cold and cloudy winter day with little sunshine and tremendous heating demand juxtaposed the relation to the factor of 25, by which the heat emission was higher than the daily insolation. In the case of London, sufficient homogeneity in the heat emission might yield to correct models and temperature forecast, as the fundamental works of Oke for the US or Santamouris for the city of Athens have shown (Oke, 1973, 1982; Santamouris, 2001). The link between insolation and anthropogenic heat has been studied on different scales, starting at the street or urban canyon (Neto & Fiorelli, 2008), observing several building blocks (Rode et al., 2014; Dong, Cao, & Lee, 2005a; Dong, Lee, & Sapar, 2005b), and evaluating large urban areas (Sailor & Lu, 2004). Measuring the influence of roofs and streets on urban climate is done by remote sensing (Li et al., 2013; Duda & Abrams, 2012), which presumes a correlation to the surface temperature. With regard to Singapore—a city of high density living heterogeneously intermixed with green spaces (parks and water catchment areas) and commercial or industrial sites—it might be questionable whether some of the assumptions of urban microclimatology are valid (Erell, Pearlmutter, & Williamson, 2012). Erell et al. (2012) describe this understanding of the standard model by (p. 54) QF is small compared to irradiation, (p. 56) a scale of below 100 m would be needed to account for intra-urban variations, (p. 72) the urban form consists of nearly continuous build facades, (pp. 76–77) a tremendous misunderstanding of urban daytime cool islands, which according

to the authors are unlikely in the tropics and not possible in conjunction with high dense built-up area with high anthropogenic heat release—which is contrary to the findings of Roth (2007), Quah and Roth (2012), Li et al. (2013), Roth and Chow (2012), Chow and Roth (2006) for Singapore, which shows an explicit urban cool island during daytime in the CBD. Convection might be the key explanatory factor for these cool islands. Urban canyon and canopy models implicitly refer to horizontal surfaces and their short- and long-wave radiative properties which are clearly a dominant feature in cities of the street-dominated urban sprawl in the US or the uniformity of roof top levels and shape in most European cities. Architects call this feature the roofscape or sea of roofs, as obvious in Fig. 1 for Athens. Another feature of microclimatology is outdoor thermal comfort (McGregor & Nieuwolt, 1998). It combines a CFD model of the urban canyon with qualitative indicators, where wind speed and direction in conjunction with building shape predominantly determine the simulation result (Moonen, Defraeye, Dorer, Blocken, & Carmeliet, 2012). An urban porous media model with the same scope though avoiding CFD is presented in Hu et al. (2012), assuming homogeneity in the urban morphology. Wind tunnel experiments and flow simulations help estimating the impact of the built environment here—yet the average wind speed in Singapore is low and consequently design is focusing on shading and greenery (Tay, 1989; Konya, 1980). The relationship between urban form and building electricity consumption is therefore of interest, as a matter of scale, resolution and magnitude (Rode et al., 2014; Ko, 2012; Futcher, Kershaw, & Mills, 2013); our investigation deals with two main constraints, one being insufficient data available on electricity consumption and the other is the tropical climate of Singapore, which might result in higher dependencies of UHI intensity on anthropogenic heat. Also we are not able to answer how the current microclimate modeling approach does depend on the morphologies of their cities of origin, we are able to show by existing measurements of Roth et al. and by our calculations of QB inconsistencies in the current approach. 1.2. Case study Singapore Currently, Singapore is a vibrant tropical megacity of 5.4 million inhabitants. The electricity consumption in the residential and commercial building sector amounts to approximately 16.7% of the total energy demand in Singapore (IEA, 2012). To maintain productivity in spite of shifting demographics and to sustain economic growth are key priorities on the government’s agenda. Limited land available for future developments and a prospected population of

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

6.5 to 6.9 million by 2030 puts additional stress on the island (NPTD, 2013). The high-dense urban living in Singapore expects urban design and planning strategies to pay special attention to the tropical climate, typically addressing high relative humidity, high day and nighttime temperatures, and low average wind speed. The hot humid weather reduces the outdoor thermal comfort and leads to increased energy demand for ventilation and air conditioning as a secondary effect. Long term climatic changes like global warming and seasonal phenomena like haze are another driver for the desire to study the city’s climate. Policymaking in urban design and planning recommends – based on microclimate studies in situ and previous knowledge on phenomena alike – a certain set of rules to be applied in Singapore (e.g., Tay, 1989; Wong et al., 2007). When dealing with cause and consequence of the local climate, much relies on how the climate is modeled and what the input parameters are. Sun and irradiance are the main drivers behind the diurnal cycle, and urban structures such as streets and buildings deeply influence the energy balance and storage equations (Oke, 1973, 1982; Mirzaei & Haghighat, 2010; Oikonomou et al., 2012). As it has been shown for Singapore before, the spatial inhomogeneity of anthropogenic heat emissions from individual high-rise buildings and their absolute contribution compared to natural insolation challenges models and theories developed under different climatic conditions and urban morphologies (Berger, 2012). Potential consequences on outdoor thermal comfort, urban microclimate and UHI are therefore a matter of anthropogenic heat distribution and scale.

1.3. Bottom-up versus top-down energy demand estimation Anthropogenic waste heat emission happens due to energy demand. For the estimation of the energy demand two main approaches can be distinguished: top-down and bottom-up. The latter calculates the energy use intensities (EUI) for representatives of buildings (Kavgic et al., 2010). An example of employing bottomup modeling for the estimation of UHI is given by Heiple and Sailor (2008). As bottom-up models require a multitude of parameter and values, the model results should be calibrated with available measured values of electricity demand. Especially if energy is calculated for larger groups of buildings, a systematic error, i.e., the difference between the calculated and the averaged measurable energy consumed per building, becomes dominant (Böhme & Hamacher, 2013). Top-down approaches depend on the availability of measured energy demand values, see the introduction to the topic in PérezLombard, Ortiz, and Pout (2008). A top-down calculation derives average EUI from aggregated energy demand and building statistics. The gross floor space (GFS) is a very common reference value for the EUI. It is widely used in architecture and real estate economy and expresses the size of a building. The energy demand of a single building or a group of buildings does not only depend on the size of the building but also on various parameters like (a) the use of the building or of a zone within the building, (b) building attributes like the age of the building, or (c) various other attributes as the social status of the building’s inhabitants. For every category, an EUI of its own has to be specified. The more categories are used for the estimation of the energy demand, the better the estimation generally is, under the prerequisite that the used attributes actually correlate with the energy demand. On the other hand, using more categories requires more information about the building itself and about the EUI of every category. A common method for gathering the EUI by category is multiple linear regression (Boulaire, Higgins, Foliente, & McNamara, 2014; Howard et al., 2012). It requires a statistical sufficient number of samples of GFS by category and energy demand. Linear regression does not need to be applied if aggregated energy

375

demand and aggregated reference values are known for the same categories. The estimation of the energy demand for single buildings based on average EUI will produce average consumption results which do not include building and user specific deviations. The actual consumption depends on various socio-economicaltechnical parameters, which are not all known for any city wide analysis. Therefore, the estimated consumption differs from the real one. The differences can reach up to a factor of ten (Ko, 2012) and lead to statistical errors when averaged values are applied to single buildings (Böhme & Hamacher, 2013). While modeling approaches for the top-down estimation of energy demand are well investigated as described before, the main problem in most cases is the acquisition of appropriate data. (Boulaire et al. (2014) and Howard et al. (2012) benefit of existing data from authorities which provide crucial parameters, for example GFS by type of use in high spatial resolution (tax slots). 2. Methods 2.1. Calculation overview An overview of our suggested combination of bottom-up and top-down calculation is given in Fig. 2. Rectangles represent data, rounded rectangles represent calculation processes. The main information inputs are georeferenced data sources. They contain spatially referenced polygons with parameters of land use and building density for over 100,000 land parcels, covering the whole of Singapore, and about 160,000 building polygons. Information about the building density is not complete, which makes assumptions necessary. Missing values are filled up regarding the number of floors. Based on this initial information, the GFS is calculated and a detailed type of use for every building is identified. The results are saved in a building database. For the case of buildings with mixed use, buildings are divided into different zones. Their size is estimated as a share of the GFS of the whole building. The given building density is a planning parameter and may vary from the actual density. Therefore, an extensive number of control samples had been generated by manually counting the number of floors of individual buildings. We corrected and calibrated the GFS, comparing this validated source with the calculation results based on planning parameters. Corrected values and their origins were stored in the building database. Sample checks were concentrated on important use types, for example the large public housing sector with high rise buildings. From all available sources, we integrated the use type descriptions and merged them into one typology. This typology represents the most detailed description of different building uses which can be derived considering the given input data. The developed method supports the addition of further use information from other sources, which may be discovered in the future. The EUI is a quotient of measured aggregated electricity consumption and the derived GFS. Therefore, the use types of the detailed use typology are combined into sectors and the electricity consumption information is categorized. The last step is the estimation of the average electricity consumption of buildings for any selection of buildings in any geographic scale based on the EUI and the building information. Output values are averaged as mean annual values of heat flux densities in W/m2 , as a power density. 2.2. Input data Our investigation bases on two major geographic information system (GIS) data sources: land use information from the Master Plan 2008 by the Urban Redevelopment Authority (URA) and

376

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

Fig. 2. Block diagram of calculation and validation steps; rectangles represent data; rounded rectangles represent calculation processes.

Fig. 3. Exemplary visualization of major data sources. Map data© 2008 Singapore Land Authority.

building footprints from the Street Directory of the Singapore Land Authority (SLA) (Hanif, 2013; EMA, 2012). Fig. 3 explains the data on the basis of an exemplary section of Singapore. Small polygons represent buildings with a building use type. Larger polygons represent land use units with a use type and a use density. The land use density is given by the gross plot ratio (GPR), that is the ratio between the GFS of the buildings within a land parcel and the footprint area of the land parcel itself. For about half of the land parcels, no numeric land use densities but only three descriptive categories are given. Information of buildings and land parcels is joined by their geographic position. Both datasets were last revised in 2008. Table 1 lists all data sources, attributes for the use type identification and the number of categories an attribute occurs in. By comparison, Fig. 3 indicates four building use types out of twelve, six land use types out of 33 and four land use densities out of 114. The electricity consumption data is taken from the Singapore Energy Statistics by the Energy Market Authority (EMA), divided into five categories. There are some “typical” Singaporean use types, which have to be explained: (1) The acronym “HDB” stands for “Housing

Development Board” and describes mostly high rise public housing projects, where stricter rules for the property and the acquisition apply. About 85% of the Singaporean population lives there; (2) “Landed Housing” describes detached, semi-detached or terraced houses with similar building heights; (3) “Condominiums” are buildings of middle and upper class standard of living. Most of them have a shared pool and are guarded. For areas without a numerical GPR value, we assumed the number of floors based on control samples. Table 2 lists the assumed

Table 1 Number of categories of attributes included to specify the use type of all buildings. Source

Attribute

Street Directory/SLA (http, 2015b) Master Plan 2008/URA (http, 2015a)

Building use

12

Land use type

33

Singapore Energy Statistics 2012/EMA (http, 2015d)

Land use density Power consumption Sectors

Number categories

114 5

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

377

Table 2 Assumptions of number of floors by GPR attribute. Area attribute

No. of land parcels

Assumed no. of floors

Justification

Landed Housing

∼72,000

2

Evaluation

∼19,500

4

A control sample in two areas with landed housing shows an average of two floors. If buildings have more floors, often terraces have to be subtracted Shows a distribution of the types of use the label evaluation is connected with. Those are mainly Standard and Industrial Buildings which are not considered to be high rise buildings. Four seems like an acceptable average, see Fig. 4 Street block plans define complex construction parameters for a group of buildings. Those are predominantly older and smaller buildings No assumptions are necessary as the given numerical value can be used

Street Block Plan GPR-value

∼1600

2.5

∼68,400

Nil

number of floors and a justification for three descriptive attributes of building density. No assumptions were necessary for about 68,000 land parcels with a numerical GPR value.

use of the building. As this building is located on a land parcel with land use type Education, we can derive that the building is used as educational institution. Therefore we conflate all information about the building use, the land use and the land use density to derive a most detailed description of building use. Based on this, we summarize multiple use categories into the five power consumption sectors, where electricity consumption data is available. We developed a concept of stepwise conflation of two use types of different typologies into a target typology in order to systemize the conflation of all information. This enables a standardized data structure which can be used for all conflation steps. It can further be extended without changing the structure itself if new information and further typologies are added from sources not yet known. Table 3 shows an example. In a first step (rule 1), type Standard and type Education are merged to the type Educational Institution. This is transformed in rule 2 to the sector Commercial/Institution. The second type of rule 2 uses a dummy type “nn” without effect, describing a transformation as a special type of conflation. All 114 categories of land use density were reduced to eight classes. Based on the conflation of land use types and density classes, both from the Master Plan 2008, a joint typology is created. This is then conflated with the building use information from the Street Directory. This leads to a joint typology with 26 use types, which are then combined to the sectors the electricity consumption data is available for. In order to get an overview of the significance of single use types of the available sources and links between them, we use a modified Sankey diagram. In this case, it may not be understood as a flow diagram but rather as a four dimensional visualization (see Fig. 4). Every vertical column with colored squares reflects the types on one typology. The height of the squares represents the GFS of that type. The grey lines show which types of different typologies occur in common. The height of the grey lines weights the occurrences by the size of the affected GFS. As an example, the upper grey band shows that buildings which (a) stand on land parcels “Very High Density” with a GPR of 2.8 or higher and (b) with residential use type are mainly marked as HDB buildings and therefore assigned to the sector Residential: HDB Building. The height of the gray band indicates the importance of HDB buildings in the Singaporean building landscape. HDB buildings sum up to almost 140 million m2 gfs in our calculations. As another example, the context of the descriptive land use density categories ‘landed housing’ and ‘evaluation’ can be identified. Landed housing is exclusively combined with residential use and residential buildings. In contrast, areas marked as evaluation are used very diversely and most of the time related to categories of the commercial sector. In order to avoid too many

2.3. Calculation of gross floor space The calculation of the GFS mainly bases on the building density information described by the GPR value of the Master Plan 2008. Assuming that all buildings of a land parcel lp have equal height, the GFS of one building i ∈ lp is calculated as shown in Eq. (1): The total GFS per land parcel is multiplied by the portion of the foot print Fi per single building divided by the sum of building footprints Fj within that land parcel. The footprint area of a building Fi [m2 ] corresponds to the portion of the earth’s surface the building covers. GFSi = Flp · GPRlp ·

F

i

(1) Fj

j∈lp

The number of floors fi of a building i can be calculated as its GFS divided by its foot print area. In order to obtain an integer, 0.5 is added to fi and the result is rounded down. Using Eq. (1), fi is given by

⎡

fi =

 GFS Fi

i



⎤

⎢ Flp · GPRlp ⎥ + 0.5⎥ ⎣  ⎦

+ 0.5 = ⎢

(2)

Fj

j∈lp

The GFS of buildings on those land parcels with descriptive building density is the product of the assumed number of floors and the building footprints as given in Eq. (3). The assumptions and justifications are listed in Table 1. GFSi = fi · Fi

(3)

2.4. Identification of building use based on multiple sources Information about the use of buildings is expressed as categorical variables (see Fig. 3). Every category defines a use type. All use types of one data source form a typology. When different data sources with varying typologies are conflated, their use types have to be compared to each other and merged into a single typology. The aim of the conflation is to derive a typology, which distinguishes the use of buildings with regard to their energy demand. Considering this, the information content of the use types varies. The building type Standard does not imply information about the Table 3 Examples for the conflation of use types from different typologies into a target typology. Conversion type

Conversion rule

Type of original typology 1

Conflation Transformation

#1 #2

Standard Educational institution

Type of original typology 2 + +

Education nn

Type of target typology → →

Educational institution Commercial/institution

378

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

Fig. 4. Correlation of use types of four typologies. Every column represents one typology. The bars’ thickness represents the related GFS. Only combinations >1 km2 GFS are illustrated.

categories presented in Fig. 4, we only show common occurrences (grey bands) exceeding one million m2 , i.e., 1 km2 . 2.5. Verification and correction of calculated GFS GPR-values of the Master Plan 2008 are planning parameters and differ from the real situation. Reasons for the difference are that buildings either have been constructed under prior regulatory conditions or due to exceptions and further regulations for the building density. We use plausibility checks, random control samples and crowd based internet sources to compare our calculations with the real situation. Corrected values are written into the building database. First of all, we only consider land parcels for the calculation of the GFS when actually a building polygon is inside. If this condition is met we check the calculated number of floors. In Singapore, the highest buildings are in the central business district (CBD) and have 73 floors. The highest HDB building has 50 floors. Therefore all buildings with more floors were filtered out and checked manually. About 500 outliers had an impact of about 5% to the overall GFS. The most frequent reason for those outliers is small building polygons on large land parcels, which leads to a high GFS and high number of floors. For about 1500 buildings, we verify the calculated number of floors by counting the floors manually based on site visits. The buildings are spread across Singapore and all sectors, but our main focus were HDB buildings since they form the most important category as they cover about 36% of the overall GFS and are built mainly on land parcels with very high building density (see Fig. 4). In order to compare calculated and checked values, we define the ratio Rlp between the GFScheck , which is calculated based on the manually counted number of floors, and the GFSGPR which is based on the building density.



fi · Fi

i∈lp GFScheck = Rlp = GFSGPR GPRlp · Flp

(4)

2.6. Electricity consumption and sensible heat flux density We calculate the building related anthropogenic heat flux density based on the average electricity load per building. The total electricity load depends on the size of GFS of the building and the EUI which depends only on the use category.

The EUI itself is calculated by the division of the aggregated electricity consumption (in TW h) of Singapore in 2009 as published by the EMA and the GFS, aggregated for all buildings (Salamanca et al., 2013). The electricity data of EMA shown in Fig. 6 is published for the sectoral use types. Eq. (5) is the calculation of the EUI per m2 GFS, EUIGFS (u). EUIGFS (u) =

E(u)



(5)

GFSi (u)

i

EUIUS (u) =

E(u)

0.8 ·



(6)

GFSi (u)

i

For comparison to other sources, we also specify the EUIUS (u) in Eq. (6) as the energy use intensity per square meter usage space (US). The usage space corresponds to the living area in the residential sector. It is the GFS without stairways, corridors and construction area like walls. Usage area is calculated as an average of 80% of the GFS. The results of both calculations are displayed in Table 4. Residential use has an EUIGFS between 30 kW h/m2 /a and 52 kW h/m2 /a. HDB Buildings have the lowest EUI of the residential sector. The EUIGFS of non-residential use is almost five times higher—between 143 kW h/m2 /a for the category Others and 172 kW h/m2 /a for the category Commerce & Service. One reason for the higher EUI of nonresidential use is the extensive use of air condition in those sectors. The EUI of Industry is the highest. Since we did not correct and check the calculated GFS of the sector Industry, the actual EUI might differ from the shown result. For the sector Transport, no EUI is calculated as the majority of the electricity consumption is not due to buildings. In 2011, Chew presented an EUI for shopping malls of about 350 kW h/m2 /a and for tertiary institutions of about 150 kW h/m2 /a (http, 2015e). Those are subcategories of the sector Commerce & Services and confirm our average result of 215 kW h/m2 US /a for this sector. According to Chew, the average EUI of the residential sector is 40–45 kW h/m2 /a. The electricity consumption per building and sector is calculated by multiplying the EUI and GFS for every building. Therefore, the applied method is a simple disaggregation of the published electricity values. If more information was available – for example the electricity consumption by planning area or by even higher geographic resolution – linear regression methods could be applied. Examples for those methods are given by Howard et al. (2012) for

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

379

Table 4 Energy use intensities by sector, distinguishing GFS and US as reference spaces. Use type

GFS [m2 ]

Total consumption [GW h/a]

EUIGF [kW h/m2 gfs /a]

EUIUS [kW h/m2 us /a]

Commerce & services Res.: condominium Res.: HDB building Res.: landed housing Industry No consumption Others Transport

82,845,900 48,747,930 138,694,305 19,290,417 80,965,266 672,912 3609,149 1260,229

14,296 1507 3902 1006 14,727 – 519 1653

172.56 30.91 28.13 52.15 181.89 – 143.80

215.70 38.64 35.16 65.19 227.37 – 179.75

the city of New York or Boulaire et al. (2014) for a district level in Australia. The climatic conditions in Singapore throughout the year are quite constant. The daily mean temperature ranges between 26 ◦ C and 28 ◦ C. Regarding precipitation and wind, differences between the rainy season and dry season can be identified (Berger, 2012), but nevertheless, they are small in comparison to seasonal changes in non-tropical regions (Chow & Roth, 2006). Therefore, the climatic influence on the power consumption is insignificant for this research. Daily activities show a higher impact on daily load profiles. Quah and Roth published load profiles derived by single institutions (Quah & Roth, 2012). For residential use, they found peak load at about midnight. A disaggregation of the total Singaporean electricity profile into sectoral load profiles revealed residential peak load at about 8 pm (Hanif, 2013). In both sources, the commercial peak load is at about 11 am (Quah & Roth, 2012; Hanif, 2013). The factor between the maximum and minimum power load per day is about two for residential apartments and between two and three for commercial use. We do not consider daily electricity load profiles on a building level, as it would add another source of uncertainty to the results. Daily profiles would be most helpful for comparing Q* with QF . Daily electricity demand profiles could be obtained by using sector-based general load profiles, or on a domestic level using household-type based profiles which are reflecting the different magnitude of HVAC use. Next, in Eq. (7), we calculate the average electricity load e [kW h] per building i as the sum over all building zones of use u.

ei =

 u

heat flux density is an indicator which combines the EUI with the building density. qi =

ei Fi

(8)



qc =

ei

i∈c

Fc

(9)

Due to limited air exchange rates of buildings, there is a time lag between the electricity consumed indoor and the measurable sensible waste heat in the atmosphere. For the time averaged balance of the flux density, we neglect storage. Gas consumption in buildings for cooking and water heating is not considered either as a contribution to anthropogenic waste heat as it accounts for only about 5% of the electricity consumption (EMA, 2012). 3. Results We present an evaluation of the calculated GFS by different use types and look at the quality of the data we used for the computation of waste heat flux density. We show detailed results of the building specific calculation of waste heat flux for a section of Singapore and explain the Singapore specific waste heat flux distribution. In order to qualify the discussion about the size of spatial zones for the input into climatology models, aggregation of building specific results on different spatial grid sizes are compared. 3.1. Evaluation of available data and manual control samples

EUIGFS (u) · GFSi (u) ·

1 8760

(7)

The derived electricity load per building is converted to an average heat flux density per building qi [kW h/m2 ]in Eq. (8) and per grid cell qc [kW h/m2 ] in Eq. (9), related to the whole grid cell foot print Fc [m2 ]. The latter also includes all areas around the buildings. The

Fig. 5 displays the distribution of Rlp which was defined as the ratio between the GFScheck , and the GFSGPR in Eq. (4) by use type for almost 400 land parcels. Each dot represents a land parcel which may contain one or more buildings. Vertical displacements of the dots are made to better visualize the density of dots. The distribution is shown as a boxplot. Within the white box are the inner 50% of the land parcels. Thick vertical black lines dividing the white boxes represent the medians of the distributions. The horizontal lines

Fig. 5. Boxplot (black) and single representations (red) of Rlp , the ratio between GFS derived from control samples, and GFS derived from GPR-based calculations. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

380

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

Fig. 6. Gross floor space (GFS) in [million m2 ] by sector and method of data acquisition.

outside the boxes show the variability outside the upper and lower quartiles. The resulting distributions are not balanced as the logical lower boundary of the ratio is zero. We calibrate GFSGPR using the median of Rlp , the median of the distribution, instead of the average. This avoids that outliers influence the results disproportionately. For HDB Buildings, the median of Rlp is about 1, indicating no systematic deviation between the GPR based calculation and the control samples. Nevertheless, the deviation of single samples is considerable, which shows that the GPR based calculation reveals true results in general, but individual areas can differ. For Condominiums, the systematic deviation is about 25% of the calculated GFS. Although considering the deviation of single samples, almost all samples show a higher GFS than calculated. Commerce & Servicebuildings show a median of 0.95. Overall, their GFS is overestimated by the GPR based calculation. Next to the above explained outliers and sample checks, we furthermore correct our calculations with data of about 100 high rise buildings from crowd based sources as Wikipedia and Highrisebuilding.info (http, 2015c). Due to the very specific architectural design, the floor numbers of mainly commercial high rise buildings can vary in comparison to building densities. Some of those buildings are on land parcels classified as “evaluation” but do not meet the assumptions we generally applied. Fig. 6 presents the GFS aggregated over all buildings in Singapore and distinguished by eight use type sectors and the three types of calculation. About 70% of the calculated GFS is based on GPR values; 12% are manually checked by the above described control samples. The floor-assumptions have a big impact especially on the homogeneous category Landed Housing and the inhomogeneous category Commerce & Service. Buildings in Singapore have a total of 390 million m2 gfs . The 10,000 HDB buildings are mainly high rise buildings with more than eight floors and account for the biggest share of 35.5%, followed by Commerce & Service with 22.7%. In contrast to that, there are about 100,000 houses of the category Landed Housing in Singapore, although this use type has just a share of about 5% of the overall GFS. Sector Commerce & Services includes commercial spaces as well as areas for education and medical care. Sector Others covers use types like utility, special use and agriculture. The sector No Consumption covers buildings which are identified as beach area, or stand on open space. 3.2. Waste heat flux density of Singapore A central result of our computations is the waste heat flux density in a building specific spatial resolution. The grid cell view presents results that are area covering for the whole Singapore main island. The building specific view only refers to building footprints and does not include the surrounding areas. All values are calculated due to electricity consumption of buildings of residential

and commercial use. The average heat flux density of all residential and commercial buildings in Singapore is 27 W/m2 fp , but we observe substantial differences on sectoral and spatial scales. Fig. 7 shows the spatial distribution of heat flux densities for a section of Singapore. Every polygon represents a building; the colors represent ranges of heat flux densities. The map covers an area ranging from Chinatown in the east (a) to the west. In the north east and south east of the section, commercial buildings along Orchard Road (area (b)) and from the Chinatown district appear in red and orange. Those areas have a high building density and a high average electricity load of 20 W/m2 gfs due to predominant commercial use. This leads to the displayed very high heat flux density of over 200 W/m2 fp . In comparison to that, the average solar irradiation in Singapore is 176 W/m2 . Buildings in area (c) are mainly landed housing properties. Landed housing properties have the highest EUI of the different types of residential use, but result in a heat flux density of just about 12 W/m2 fp . Area (d) is a typical example of high rise residential buildings and smaller commercial buildings. They show a heat flux density between 20 W/m2 fp and 100 W/m2 fp and are grouped into small spatial zones of similar heat flux density, shown by equally colored regions in the map. But zones of very different levels of heat flux density are located right next to each other. In Singapore, large areas with similar characteristics are spread across the city (Tan, 1999). This can also be seen in the surroundings of area (d). Fig. 8 shows the waste heat flux density by size of earth surface, covered by building footprint area. This is an interesting aspect when we try to locate waste heat fluxes. Buildings of the sector Landed Housing have a small footprint area. In sum they contribute only 22% of the total building footprint area in Singapore, although every second building in Singapore is of this sector. In comparison, buildings of the sector Commerce & Service have a large footprint area. They cover 14% of the buildings in Singapore but 33% of the building footprint area. From Fig. 8 one can furthermore see the correlation of building use and heat flux density in Singapore. Low heat flux densities below 20 W/m2 do mainly exist for the categories Landed Housing and part of Condominiums, the private housing sector. Those are small buildings and although they have the highest EUI of the residential sector, they do not reach high heat flux densities. Heat flux densities between 20 W/m2 and 60 W/m2 appear in the high rise HDB Building sector and for high rise Condominiums. Both sectors have some very high rising buildings with heat flux densities above 60 W/m2 . Due to the high EUI of buildings in the sector Commerce & Service, we calculated heat flux densities of over 60 W/m2 for 90% of the building footprint areas in this sector. High rise commercial buildings – mainly located in the CBD – lead to the existence of very high heat flux densities of over 200 W/m2 .

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

381

Fig. 7. Heat flux density due to building electricity consumption of residential and commercial use of a section of Singapore. Map data© 2008 Singapore Land Authority. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3.3. Averaged heat flux density per grid cell Beside the building specific heat fluxes we analyze grid cell resolution with cell side length of 250 m and 1 km, similar to Li et al. (2013) and Mavrogianni et al. (2011), respectively. Fig. 9(a) and (b) display the heat flux densities on a spatially aggregated grid level for the Singapore main island Pulau Ujong. Fig. 9(a) shows a grid resolution of 1 km×1 km, Fig. 9(b) uses a resolution of 250 m×250 m. The 1 km resolution is commonly used for research on UHI effects, yet models with higher resolution are upcoming. Quah and Roth (2012) studied three sample areas of 500 m radius for their investigations. The 250 m resolution is chosen in order to compare two different aggregation levels. The legends on the right side show the color range, the number of grid cells of a range and the relative share of the total 528 and 4810 cells, respectively. The heat flux densities of Fig. 9 are related to the whole grid cell area of 1 km2 (or 1/16 km2 , resp.). That means, the reference area for the calculation of the heat flux density is much larger than just the building footprint, as areas surrounding the buildings – streets, greenery and parking lots – are considered as well. This leads to the explicitly lower densities. The ratio between the sum of building footprints within a grid cell and the grid cell footprint varies between 0% and 32.49%. There is a strong correlation between the filling ratio of a grid cell and the calculated heat flux density. For 1 km grid cells with a heat flux density between 0.5 and 3.1 W/m2 , it is 3%, for cells in the range of 12–20 W/m2 , it is 16.5%, and for

the two cells in the range 60–100 W/m2 , it is above 30%. 75% of the cells in Fig. 9(a) shows a density of less than 6.3 W/m2 . The averaging effect is increasing with lower spatial resolution. While still 0.4% of the grid cells in Fig. 9(b) are within the two highest heat flux ranges, no grid cell of Fig. 9(a) shows a higher average heat flux than 100 W/m2 . On the west side of Singapore, large areas of industrial use exist. As data availability does not allow consideration of the industrial sector, it does not show up in the figures. 3.4. Indications of inhomogeneity Fig. 9(a) is averaging the heat flux densities from buildings’ electricity consumption in the residential and commercial sector only, and the visualization based on 1 km grid size does not indicate any special inhomogeneity. On the contrary, the assembly shows peak values in the CBD and lower values towards the boundaries. Heat from other sources and sectors (transportation, industry, etc.) would produce different patterns of distribution. All sector-based maps would have to be summed up in order to provide the full picture of anthropogenic heat emissions. In Fig. 10(a), the average heat flux density of Fig. 9(a) is only related to the footprint area of the buildings, not to the whole grid area. The peak values rise significantly and indicate by more warm colors that certain cells contain one or more outliers in terms of high energy flux per footprint. Fig. 10(b) combines the building view of Fig. 7 with both kinds of averaging from Figs. 9(a) and 10(a). The

Fig. 8. Heat flux distribution by use sector and building footprint area.

382

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

Fig. 9. Average heat flux densities due to building electricity consumption of the residential and commercial sector in (a) spatial resolution of a 1 km grid and (b) spatial resolution of a 250 m grid.

central cell contains mostly empty area with zero anthropogenic heat from buildings, several low- and medium rise residential buildings with moderate heat flux below 20 W/m2 , and few commercial buildings with emission equal or beyond the irradiance on average time and footprint. This representation now emphasizes buildings which distort the illusion of a single, homogenous roofscape as the one shown in Fig. 1. If there was only little distortion or if only a limited number of cells within the whole city were showing outliers, the microclimatic simulation of the UHI would not be compromised. As Fig. 10 suggests a comparable high level of inhomogeneity, it might be questionable if standard assumptions

for the urban canopy and land use model are suitable. Li et al. (2013) apply a single set of building dimensions for each of the three subcategories, namely commercial/industrial, high intensity residential and low intensity residential, with 113, 18 and 13 W/m2 heat emissions, respectively. 4. Discussion The building data we used stems from the Urban Redevelopment Authority, responsible for the city’s future development. The parameters are guidelines for future construction and do not need

Fig. 10. Average heat flux densities due to building electricity consumption of the residential and commercial sector in spatial resolution of 1 km; density reference parameter is the building footprint area within a grid cell. Map data© 2008 Singapore Land Authority.

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

to reflect the actual situation. Therefore we used manually generated control samples to test how well those parameters describe the actual situation. The control samples showed that the planning parameters lead to an underestimation of building densities amounting to 40% for buildings in the private housing sector. The building density of the public housing sector is well described and the density of commercial buildings is overestimated by 5%. The actual situation regarding building density is a lot more heterogeneous than planning parameters represent. This will lead to a higher error of the estimated energy demand. Nevertheless the data used is the only data available for an area covering estimation of building densities and use types. It enables an estimation of building based energy demand with acceptable effort. The average waste heat flux density of all residential and commercial buildings in Singapore is 27 W/m2 , but we observe substantial differences on sectoral and spatial scales. Heat flux densities were found in between 3.1 and 1500 W/m2 . Areas with Landed Housing indicate homogeneous heat flux densities of lower than 20 W/m2 . In areas of mixed commercial and high density residential use heat fluxes vary by the factor of 10 between 20 W/m2 and 200 W/m2 due to the different use types and varying building heights. Caused by the limitation in the available data, various socio-economic parameters that influence electricity consumption are not considered. Especially for the high rising building sector, the differences in electricity consumption will average out due to the large number of units within one building. The results can be used for further investigations in the fields of urban climate and energy demand, like modeling the electricity grid with high spatial resolution. Once buildings or precincts with high average heat flux density as in Figs. 7 and 10 have been identified, addressing the temporal resolution should be done by on-site measurements of the electricity consumption, either directly (Berger, 2011) or indirectly (Salamanca et al., 2013).

5. Conclusion This paper presented a general workaround for estimating building-based energy demand by combining a top-down approach on sectoral energy demand with a bottom-up GIS derived building database. The building database was created combining different GIS-datasets and focuses on building size and type of use. The methodology can be applied for cities where energy demand data is insufficient. Large scale maps can easily be generated and used for consecutive studies. As exemplary shown for the residential and commercial building sector of Singapore which represents about 16.7% of the total city-wide energy demand, the patterns of the energy flux indicate much larger inhomogeneities comparing to traditional case study cities like London (Dong et al., 2005a). Landed Housing results to homogenous and low heat flux densities. Urban climate models for this kind of land use are valid in tropical and high-dense cities. Due to coarse zoning and clustering of similar use types, inhomogeneity in the energy flux density is found across the zone boundaries. Especially buildings in the sector Commerce & Service show significant higher heat flux densities, with a majority above 60 W/m2 . Prominent features of inhomogeneity are: the aspect ratio of building height to roof surface with a pronounced verticality, limited horizontal roof surfaces, hence less radiative cooling, peak magnitudes of individual buildings’ anthropogenic heat emissions from electricity consumption in residential and commercial use, and the scale of the urban canyon. Our workaround and the results have consequences on modeling the UHI: The warm-wet tropical climate of Singapore has higher diurnal than annual temperature cycles, “a monotonous similarity of temperature” (Konya, 1980) with low average wind speeds.

383

Current micrometeorological models(including urban canyons), which assume certain levels of homogeneity and wind phenomena in the planetary boundary layer (PBL),might not be applicable in tropical and high-dense cities. The workaround enables UHI research to identify critical grid cells in simulations, where CFD could be applied for better accuracy. Acknowledgement This work was financially supported by the Singapore National Research Foundation under its Campus for Research Excellence And Technological Enterprise (CREATE) Program. Moreover, the Urban Redevelopment Authority and the Singapore Land Authority supplied important data. References Berger, M. (2011). Validierung des Energy Hub Konzepts mittels einer Fallstudie in Baden-Dättwil: Nr. 19707. Zürich: Eidgenössische Technische Hochschule ETH Zürich (Diss.). Berger, M. (2012). Urban heat-balling—A review of measures on reducing heat in tropical and subtropical cities. In T. K. Sheng, & P. Hing (Eds.), Sustainable future energy 2012 & 10th sustainable energy and environment (pp. 445–451). Brunei: Darussalam. Böhme, P., & Hamacher, T. (2013). Analyzing the errors of heat consumption estimations at high spatial resolution. In G. Hauser, T. Lützkendorf, & N. Eßig (Eds.), sb13 Munich implementing sustainability—Barriers and chances (pp. 185–194). Munich: Fraunhofer IRB Verlag. Boulaire, F., Higgins, A., Foliente, G., & McNamara, C. (2014). Statistical modelling of district-level residential electricity use in NSW, Australia. Sustainability Science, 9, 77–88. Chow, W. T. L., & Roth, M. (2006). Temporal dynamics of the urban heat island of Singapore. International Journal of Climatology, 26, 2243–2260. Dong, B., Cao, C., & Lee, S. E. (2005). Applying support vector machines to predict building energy consumption in tropical region. Energy and Buildings, 37, 545–553. Dong, B., Lee, S. E., & Sapar, M. H. (2005). A holistic utility bill analysis method for baselining whole commercial building energy consumption in Singapore. Energy and Buildings, 37, 167–174. Duda, K. A., & Abrams, M. (2012). ASTER satellite observations for international disaster management. Proceedings of the IEEE, 100, 2798–2811. EMA. (2012). Singapore energy statistics 2012: Energising our nation. Singapore: Energy Market Authority. Erell, E., & Williamson, T. (2006). Simulating air temperature in an urban street canyon in all weather conditions using measured data at a reference meteorological station. International Journal of Climatology, 26, 1671–1694. Erell, E., Pearlmutter, D., & Williamson, T. (2012). Urban microclimate: Designing the spaces between buildings:. New York: Routledge. Futcher, J. A., Kershaw, T., & Mills, G. (2013). Urban form and function as building performance parameters. Building and Environment, 62, 112–123. Hamilton, I. G., Davies, M., Steadman, P., Stone, A., Ridley, I., & Evans, S. (2009). The significance of the anthropogenic heat emissions of London’s buildings: A comparison against captured shortwave solar radiation. Building and Environment, 44, 807–817. Hanif, S. (2013). Spatial disaggregation of electric loads in megacities. München: TUM. Heiple, S., & Sailor, D. J. (2008). Using building energy simulation and geospatial modeling techniques to determine high resolution building sector energy consumption profiles. Energy and Buildings, 40, 1426–1436. Howard, B., Parshall, L., Thompson, J., Hammer, S., Dickinson, J., & Modi, V. (2012). Spatial distribution of urban building energy consumption by end use. Energy and Buildings, 45, 141–151. Hu, Z., Yu, B., Chen, Z., Li, T., & Liu, M. (2012). Numerical investigation on the urban heat island in an entire city with an urban porous media model. Atmospheric Environment, 47, 509–518. (2015). http://www.ura.gov.sg/uol/master-plan/View-Master-Plan/master-plan2008.aspx [accessed: May 2015]. (2015b). http://www.sla.gov.sg [accessed: May 2015]. (2015c). http://en.wikipedia.org/wiki/List of tallest buildings in Singapore [accessed: May 2015]. (2015d). https://www.ema.gov.sg/ [accessed: May 2015]. (2015e). http://www.schneider-electric.com.sg/documents/company/event/Inspiration 2011/T3-3-Green-Building.pdf [accessed: May 2015]. IEA. (2012). Energy statistics of non-OECD countries. Paris: International Energy Agency. Kavgic, M., Mavrogianni, A., Mumovic, D., Summerfield, A., Stevanovic, Z., & DjurovicPetrovic, M. (2010). A review of bottom-up building stock models for energy consumption in the residential sector. Building and Environment, 45, 1683–1697. Ko, Y. (2012). Urban form and residential energy use: A review of design principles and research findings. Journal of Planning Literature, 57, 301–316. http://dx.doi. org/10.1016/j.atmosenv.2012.04.024

384

P. Boehme et al. / Sustainable Cities and Society 19 (2015) 373–384

Kolokotroni, M., Zhang, Y., & Watkins, R. (2007). The London Heat Island and building cooling design. Solar Energy, 81, 102–110. Kolokotroni, M., Davies, M., Croxford, B., Bhuiyan, S., & Mavrogianni, A. (2010). A validated methodology for the prediction of heating and cooling energy demand for buildings within the urban heat island: Case-study of London. Solar Energy, 84, 2246–2255. Konya, A. (1980). Design primer for hot climates. Architectural Press Ltd. Kotthaus, S., & Grimmond, C. (2012). Identification of micro-scale anthropogenic CO2 , heat and moisture sources—Processing eddy covariance fluxes for a dense urban environment. Atmospheric Environment, 57, 301–316. Li, X.-X., Koh, T.-Y., Entekhabi, D., Roth, M., Panda, J., & Norford, L. K. (2013). A multiresolution ensemble study of a tropical urban environment and its interactions with the background regional atmosphere. Journal of Geophysical Research: Atmospheres, 118, 9804–9818. Mavrogianni, A., Davies, M., Batty, M., Belcher, S., Bohnenstengel, S., Carruthers, D., et al. (2011). The comfort, energy and health implications of London’s urban heat island. Building Services Engineering Research and Technology, 32, 35–52. McGregor, G. R., & Nieuwolt, S. (1998). Tropical climatology: An introduction to the climates of the low latitudes. John Wiley & Sons Ltd. Mirzaei, P. A., & Haghighat, F. (2010). Approaches to study urban heat island—Abilities and limitations. Building and Environment, 45, 2192–2201. Moonen, P., Defraeye, T., Dorer, V., Blocken, B., & Carmeliet, J. (2012). Urban physics: Effect of the micro-climate on comfort, health and energy demand. Frontiers of Architectural Research, 1, 197–228. Neto, A. H., & Fiorelli, F. A. S. (2008). Comparison between detailed model simulation and artificial neural network for forecasting building energy consumption. Energy and Buildings, 40, 2169–2176. National Population and Talent Division. (2013). A sustainable population for a dynamic Singapore—Population white paper. In Office P.M.s., editor. NPTD. Oikonomou, E., Davies, M., Mavrogianni, A., Biddulph, P., Wilkinson, P., & Kolokotroni, M. (2012). Modelling the relative importance of the urban heat island and the thermal quality of dwellings for overheating in London. Building and Environment, 57, 223–238.

Oke, T. R. (1973). City size and the urban heat island. Atmospheric Environment, 7, 769–779 (1967). Oke, T. R. (1982). The energetic basis of the urban heat island. Quarterly Journal of the Royal Meteorological Society, 108, 1–24. Pérez-Lombard, L., Ortiz, J., & Pout, C. (2008). A review on buildings energy consumption information. Energy and Buildings, 40, 394–398. Quah, A. K., & Roth, M. (2012). Diurnal and weekly variation of anthropogenic heat emissions in a tropical city, Singapore. Atmospheric Environment, 46, 92–103. Rode, P., Keim, C., Robazza, G., Viejo, P., & Schofield, J. (2014). Cities and energy: Urban morphology and residential heat energy demand. Environment and Planning B: Planning and Design, 41, 138–162. Roth, M., & Chow, W. T. L. (2012). A historical review and assessment of urban heat island research in Singapore. Singapore Journal of Tropical Geography, 33, 381–397. Roth, M. (2007). Review of urban climate research in (sub) tropical regions. International Journal of Climatology, 27, 1859–1873. Sailor, D. J., & Lu, L. (2004). A top-down methodology for developing diurnal and seasonal anthropogenic heating profiles for urban areas. Atmospheric Environment, 38, 2737–2748. Salamanca, F., Georgescu, M., Mahalov, A., Moustaoui, M., Wang, M., & Svoma, B. (2013). Assessing summertime urban air conditioning consumption in a semiarid? Environmental Research Letters, 8, 034022. Santamouris, M. (2001). Energy and climate in the urban built environment. London, UK: James & James. Tan, S. (1999). Home, work, play. Urban Redevelopment Authority. Tay, K. S. (1989). Mega-cities in the tropics: Towards an architectural agenda for the future. Institute of Southeast Asian Studies. Wong, N. H., Kardinal Jusuf, S., Aung La Win, A., Kyaw Thu, H., Syatia Negara, T., & Xuchao, W. (2007). Environmental study of the impact of greenery in an institutional campus in the tropics. Building and Environment, 42, 2949–2970. Zhou, Y., Weng, Q., Gurney, K. R., Shuai, Y., & Hu, X. (2012). Estimation of the relationship between remotely sensed anthropogenic heat discharge and building energy use. ISPRS Journal of Photogrammetry and Remote Sensing, 67, 65–72.